diff --git a/README.md b/README.md
index a72b31d8..104b0f08 100644
--- a/README.md
+++ b/README.md
@@ -31,6 +31,11 @@ is built with an ancient wasi-sdk to workaround
 * Correctness
 * Clean code
 * Small footprint
+* Portability
+  * Core wasm: C11
+  * WASI: POSIX-like
+  * Clang/LLVM extentions are ok where an alternative implmentation
+    in pure C is reasonably easy.
 
 ## Non-goals
 
@@ -72,42 +77,19 @@ It includes
 
 ## Where can this run?
 
-### The most tested
-
-* macOS/amd64
-* wasm32-wasi (on toywasm itself)
-
-### Tested on CI
-
-#### Natively
-
-* macOS/amd64
-* Ubuntu/amd64
-
-#### With qemu
-
-* Ubuntu/arm64
-* Ubuntu/armhf (Note: 32-bit)
-
-#### With qemu (less coverage because of slowness)
-
-* Ubuntu/s390x (Note: big endian)
-* Ubuntu/risc-v
-
-(I haven't investigated why they are slower than others.
-It might be related to the fact that our build disables LTO
-for them for toolchain issues.)
-
-#### With toywasm (less coverage because of slowness)
-
-* wasm32-wasi
-
-### Occasionally tested manually
-
-* wasm32-wasi-threads
-* NuttX/xtensa
-* NuttX/sim on macOS/amd64
-* NetBSD/amd64
+|Platform                |Tested on CI     |Notes                        |
+| ---------------------- | --------------- | --------------------------- |
+|macOS/amd64             |Yes              |                             |
+|wasm32-wasi             |Yes (on toywasm) |                             |
+|Ubuntu/amd64            |Yes              |                             |
+|Ubuntu/arm64            |Yes (on qemu)    |                             |
+|Ubuntu/armhf            |Yes (on qemu)    |32-bit                       |
+|Ubuntu/s390x            |Yes (on qemu)    |Big endian                   |
+|Ubuntu/riscv64          |Yes (on qemu)    |                             |
+|wasm32-wasi-threads     |No               |Occasionally tested manually |
+|NuttX/esp32             |No               |Occasionally tested manually |
+|NuttX/sim on macOS/amd64|No               |Occasionally tested manually |
+|NetBSD/amd64            |No               |Occasionally tested manually |
 
 ## How slow/fast is this?
 
diff --git a/benchmark/ffmpeg.md b/benchmark/ffmpeg.md
index 52970334..8a1becab 100644
--- a/benchmark/ffmpeg.md
+++ b/benchmark/ffmpeg.md
@@ -93,6 +93,13 @@ measure the time taken.
           1637302272  peak memory footprint
 ```
 
+* Run on a macOS/amd64 laptop:
+
+  ```
+  MacBook Pro (15-inch, 2018)
+  2.2 GHz 6-Core Intel Core i7
+  ```
+
 ## Observations
 
 * toywasm performs reasonably well for an interpreter.
diff --git a/benchmark/startup.md b/benchmark/startup.md
index e35cb21a..58513668 100644
--- a/benchmark/startup.md
+++ b/benchmark/startup.md
@@ -12,7 +12,12 @@ print the `-version` message)
 
 ![Result](./startup.png)
 
-* Run on a macOS/amd64 laptop.
+* Run on a macOS/amd64 laptop:
+
+  ```
+  MacBook Pro (15-inch, 2018)
+  2.2 GHz 6-Core Intel Core i7
+  ```
 
 * [Raw values](./startup.txt)
 
diff --git a/test/run-spidermonkey.sh b/test/run-spidermonkey.sh
index 75ad0fb8..19534eb9 100755
--- a/test/run-spidermonkey.sh
+++ b/test/run-spidermonkey.sh
@@ -1,5 +1,9 @@
 #! /bin/sh
 
+# Note: spidermonkey uses nan-boxing, which is one of motivations of
+# nan-canonicalization in wasm.
+# cf. https://github.com/WebAssembly/design/issues/1463
+
 set -e
 BIN=.spidermonkey/spidermonkey.wasm
 if [ ! -f ${BIN} ]; then